How verygoodplugins/automem works
AutoMem is a long-term memory service for AI assistants. Clients submit memories through REST or MCP; the service stores them in a graph and vector index, then recalls relevant memories using hybrid search and graph relationships. Optional background enrichment and consolidation add structure and maintain memory quality, while synchronization can repair index drift. The README documents these capabilities; some implementation relationships below are based on the SOURCE INDEX, so unsampled internals are not expanded beyond supported links.
Main entry points: runtime_bootstrap.py API Registration · vector_store.py Qdrant Vectors · graph_store.py FalkorDB Graph · runtime_recall_helpers.py Hybrid Recall
Components 22
Client Interfaces
- MCP Bridge
mcp-sse-server/server.js
Memory Service
- Flask Service
app.py - API Registration
automem/api/runtime_bootstrap.py - Memory API
automem/api/memory.py - Recall API
automem/api/recall.py - Entity API
automem/api/entity.py - Graph API
automem/api/graph.py - Admin API
automem/api/admin.py - Streaming API
automem/api/stream.py - Service Configuration
automem/config.py
Memory Processing
- Embedding Providers
automem/embedding/provider_init.py - Enrichment Pipeline
automem/enrichment/runtime_orchestration.py - Memory Classification
automem/classification/memory_classifier.py - Consolidation Engine
automem/consolidation/identity_synthesis.py
Storage and Search
- FalkorDB Graph
automem/stores/graph_store.py - Qdrant Vectors
automem/stores/vector_store.py - Hybrid Recall
automem/search/runtime_recall_helpers.py - Relation Search
automem/search/runtime_relations.py - Keyword Search
automem/search/runtime_keywords.py
Maintenance
- Backup Operations
automem/backup.py - Index Sync Repair
automem/sync/runtime_worker.py
Other components
- AI Assistant
Connections 30
- AI Assistant → MCP Bridge: uses MCP
README.md - AI Assistant → Flask Service: calls REST
README.md - MCP Bridge → Flask Service: forwards requests
mcp-sse-server/server.js - Flask Service → API Registration: wires routes
app.py - API Registration → Memory API: registers
automem/api/runtime_bootstrap.py - API Registration → Recall API: registers
automem/api/runtime_bootstrap.py - API Registration → Entity API: registers
automem/api/runtime_bootstrap.py - API Registration → Graph API: registers
automem/api/runtime_bootstrap.py - API Registration → Admin API: registers
automem/api/runtime_bootstrap.py - API Registration → Streaming API: registers
automem/api/runtime_bootstrap.py - Memory API → FalkorDB Graph: stores canonical records
README.md - Memory API → Qdrant Vectors: stores embeddings
README.md - Recall API → Hybrid Recall: runs retrieval
README.md - Hybrid Recall → Qdrant Vectors: queries vectors
README.md - Hybrid Recall → FalkorDB Graph: traverses graph
README.md - Hybrid Recall → Relation Search: expands relations
- Hybrid Recall → Keyword Search: uses keyword matching
- Embedding Providers → Qdrant Vectors: provides vectors
README.md - Enrichment Pipeline → FalkorDB Graph: adds structure
README.md - Enrichment Pipeline → Qdrant Vectors: uses semantic search
README.md - Consolidation Engine → FalkorDB Graph: consolidates memories
README.md - Consolidation Engine → Qdrant Vectors: updates index
README.md - Flask Service → Memory Classification: imports classifier
app.py - Enrichment Pipeline → Service Configuration: uses settings
- Consolidation Engine → Service Configuration: uses settings
automem/consolidation/identity_synthesis.py - Backup Operations → FalkorDB Graph: backs up
README.md - Backup Operations → Qdrant Vectors: backs up
README.md - Index Sync Repair → FalkorDB Graph: checks graph records
automem/sync/runtime_worker.py - Index Sync Repair → Qdrant Vectors: checks vector records
automem/sync/runtime_worker.py - Index Sync Repair → Embedding Providers: queues repair embeddings
automem/sync/runtime_worker.py